Archi Automate User Guide

Last updated on Jun 1, 2026.

Getting started

Archi Automate brings conversational AI directly into your live Revit sessions. This part of the guide explains what the product is, how its pieces fit together, what you need to run it, and how to install and connect it to your AI client.

What is Archi Automate

Archi Automate is the AI automation bridge for Revit. It connects large language models — specifically Claude (Anthropic) and GPT/Codex (OpenAI) — to your open Revit models through the Model Context Protocol (MCP). With it, you can query, analyze, and automate your projects using plain natural language, and, where your AI client supports it, by voice.

Because the AI works against your live model, you can ask questions such as "How many rooms are on Level 2?" or "What is the total wall area in this project?" and get answers sourced directly from the model — with no scripting and no need to know Revit commands.

What sets Archi Automate apart from other AI-to-Revit tools is that it has no fixed ceiling. Instead of offering a fixed library of predefined commands, it lets the AI carry out an unrestricted range of operations against the live model. This means any operation the Revit API supports can be described in natural language and carried out. Tools built on a fixed set of commands cannot match this, because they can only do what was anticipated in advance.

Archi Automate requires a connected, MCP-compatible AI client to do anything. The bridge itself does not include a chat interface or a voice feature — the conversation, and any voice input, are provided by your AI client (for example Claude Desktop). Archi Automate is the layer that lets that AI client reach your Revit model safely.

How it works

Archi Automate uses a strictly separated three-layer design. A request flows from your AI client down to Revit and the result flows back up the same path:

1. AI agent — Claude, GPT/Codex, or any MCP-compatible client. This is where you type or speak your request.
2. MCP server — presents Archi Automate's tools to the AI agent and forwards validated requests onward. It never touches Revit directly.
3. Bridge add-in — the only component that calls the Revit API. It carries out the work, manages transactions, and enforces the safety policy.
4. Revit — your open documents in Revit 2025, 2026, or 2027.

This separation matters for safety. The AI agent never reaches Revit directly, and the MCP server never touches Revit either — only the Bridge does. Every request is checked before it runs:

• Approval for changes: all write operations require explicit approval and are governed by the active policy. Reading information from the model is always allowed; modifying it is not, unless policy permits it.
• Operation safety: before any operation runs, it passes through a safety layer that blocks dangerous actions such as file, network, and process access. This is a defense-in-depth safeguard, not a guarantee that every operation is correct — always preview unfamiliar changes first.
• Automatic rollback: write operations run inside a managed Revit transaction. If anything fails, the transaction is rolled back automatically and your model is left unchanged.
• Full audit trail: every request and its result is recorded, giving you a complete record of all AI activity against your model.

Supported Revit versions and AI clients

Archi Automate runs on Windows and supports the following Revit versions:

• Revit 2025
• Revit 2026
• Revit 2027

A version-specific Bridge is installed for each Revit version detected on your machine. You can have several Revit sessions open at once: Archi Automate detects all running sessions and the AI can switch between them, so a single AI client can work across multiple concurrent projects.

For the AI client, you can use any MCP-compatible agent, including:

• Claude Desktop
• Claude Code
• OpenAI Codex / GPT-based agents
• Any other MCP-compatible client

The quality of the AI's results depends heavily on the model tier you use. For reliable results, the recommended minimum tiers are Claude Opus 4.8 High or GPT-5.5 High (or better). Voice input is available wherever your chosen AI client provides it; it is a feature of the client, not of Archi Automate.

Installation

Installing Archi Automate adds a background Bridge to each supported Revit version, an MCP server for your AI client to connect to, and the Hub dashboard. After installing, you register the MCP server in your AI client so the two can talk.

Installing Archi Automate

Installation requires local administrator rights. Follow these steps:

1. Close Revit.
2. Run the Archi Automate installer as administrator.
3. Complete the installer pages and any required activation steps.
4. Start Revit.
5. Confirm that Archi Automate has loaded and that the Hub is reachable from the Start menu or system tray.
6. In your AI client, confirm that the MCP server is registered (see the next section).

By default, Archi Automate installs to C:\Program Files (x86)\Archigrafix\ArchiAutomate. The installer detects which supported Revit versions are present and installs the matching per-version Bridge binaries, the version-agnostic MCP server, and the Hub dashboard.

The Bridge runs as a background service and does not add a ribbon button or any UI element inside Revit. It is headless from Revit's perspective; all interaction happens through your AI client and, for monitoring and policy, through the Hub.

If the Bridge does not appear to load, the most common causes are running the installer without administrator rights, or leaving Revit open during installation. Close Revit and reinstall as administrator to resolve these.

Connecting your AI client

After installation, you must register the Archi Automate MCP server in your AI client's MCP configuration. Until you do this, the MCP server runs but no AI agent can discover it.

Claude Code. Run the following command:

claude mcp add archi-automate "C:\Program Files (x86)\Archigrafix\ArchiAutomate\mcp\ArchiAutomate.Mcp.exe"

Claude Desktop. Add ArchiAutomate.Mcp.exe as a stdio MCP server in your claude_desktop_config.json file, pointing to the same path under the install directory.

For any other MCP-compatible client, register ArchiAutomate.Mcp.exe as a stdio MCP server following that client's instructions.

The Hub dashboard displays the current connection configuration and status for reference, so you can confirm at a glance whether your client is connected.

First run and the 14-day trial

The first time Revit starts after installation, the Bridge loads automatically, validates its license state, and begins listening for your AI client — all silently in the background. The MCP server starts on demand the moment your AI client connects.

If no license key is present on first run, a 14-day full-featured trial activates automatically. No key is required to begin: every feature is available during the trial period. If the trial expires and no license is available, tool calls are refused until a valid license is activated.

License types

Archi Automate offers the following licensing options:

• Trial: a 14-day, full-featured trial that activates automatically on first run. No key is required, and the trial state is checked at startup.
• Desktop license: a single-machine activation tied to that computer's hardware fingerprint.
• Floating license: a seat checked out from your company's shared pool when the Bridge starts, and released back to the pool when it shuts down. This suits teams that share a number of seats across many machines.

Using Archi Automate

Archi Automate works through a connected MCP-compatible AI client (such as Claude Desktop, Claude Code, or another agent that speaks the Model Context Protocol). You talk to your Revit model in plain language, and the AI selects and calls the right tool on your behalf. The sections below describe what you can do, from simple questions through full model automation.

Everything in this guide assumes Revit is open with a document loaded, the Archi Automate Bridge is running in that Revit session, and the MCP server is registered in your AI client. No ribbon button is added inside Revit; the Bridge runs silently as a background service.

Ask questions in natural language

You do not need to know Revit commands or the Revit API to get information out of a model. Type a plain-language question into your AI client and the AI works out which tool to call, runs it against the live model, and answers you in natural language.

Example questions you can ask:

• "How many rooms are on Level 2?"
• "What is the total wall area in this project?"
• "List all rooms larger than 20 square metres."
• "Which walls are load-bearing?"
• "Find all rooms with no department parameter set."

Behind the scenes the AI may call element collection, parameter retrieval, or an advanced read-only operation — but you never have to choose. The answer comes straight from the open Revit document, so it always reflects the current state of the model. There is no scripting and no Revit-command knowledge required.

If you ask a question and the AI cannot find a session, ask it to list and select the running Revit sessions first (see Working with multiple Revit sessions).

Collect elements and parameters

When you want structured data about specific parts of the model, the AI uses two read-tier tools: revit_collect_elements to find the elements you mean, and revit_get_parameters to read values from them.

• Collect elements filters the model by Revit category (for example Doors, Walls, or Rooms), by Revit class name, or by filter criteria, and returns the matching element IDs together with basic metadata.
• Get parameters takes a set of element IDs and parameter names and returns the current values from the model — for example room names and areas, or the Fire Rating on a set of walls.

This is how questions such as "how many doors are on Level 1" or "list all room names and areas" are answered without any code.

Scope limit: very broad queries are subject to the maxElementsPerOperation limit defined in the active policy (5000 elements by default). This protects Revit from being overwhelmed by a query that would otherwise touch the entire model. If a query is too broad, narrow the filter — for example, restrict it to one level, one category, or a more specific condition — and ask again.

Advanced read-only operations

Some questions cannot be expressed by simple element and parameter filters. For these, the AI carries out an advanced read-only operation against the full Revit API and returns structured results. This is handled by the revit_run_readonly_operation tool, and you still describe the task in plain language.

Advanced read-only operations are well suited to tasks such as:

• Traversing host-to-hosted relationships (for example, finding every door hosted in a given wall).
• Computing derived geometry that is not stored as a parameter.
• Aggregating or cross-referencing data across linked models.
• Running model audits, such as locating elements that are missing required data.

What is guaranteed: an advanced read-only operation cannot change the model — it runs with no transaction. Dangerous actions such as file, network, and process access are blocked by the safety layer before anything runs, as a defense-in-depth check. Every operation also runs under a watchdog timeout (60 seconds by default); if it runs longer it is cancelled and an error is returned, and your Revit session stays usable.

If an operation times out, ask the AI to narrow the query or process elements in smaller batches.

Write tasks

Write tasks let the AI modify the model. As with read operations, you describe the change in plain language — but here the work runs inside a managed Revit transaction through the revit_run_write_task tool. If it fails, the transaction is rolled back automatically and the model is left unchanged.

Typical write tasks include:

• Renaming elements according to a rule (for example, prefixing every room name with its level name).
• Updating a parameter across many elements at once (for example, setting Fire Rating to "60 min" for all walls in the Core zone).
• Creating or modifying geometry, or running complex multi-element, multi-category transformations.

Write mode must be enabled in policy. Write tasks are blocked by default in readOnly mode and are rejected with a not_allowed error. Two write-capable modes are available:

• dryRun — the task runs but the transaction is always rolled back, so you can preview the planned changes without committing them.
• unrestricted — the task runs and commits, applying the change to the model.

Recommendation: for any unfamiliar operation, try dryRun first. Review the previewed changes, and only then re-run in unrestricted mode to commit. The safety layer and watchdog timeout apply to write tasks just as they do to read operations.

Note that Archi Automate does not save the Revit model after a write task completes — saving remains your responsibility, exactly as with any manual edit.

Voice input

You can interact with your Revit model by voice wherever your AI client provides a voice mode. Voice is a client feature, not something built into Archi Automate. When you speak a question or instruction, the AI client transcribes it and then follows exactly the same tool path as a typed request — there is no separate voice pipeline inside the Bridge.

To use voice:

1. Enable voice input in your AI client (for example, Claude Desktop with voice mode).
2. Speak a question or instruction directed at the model, such as "How many parking spaces are on this level?"
3. The AI acts on the transcribed request and the result is displayed or spoken back.

If a spoken element reference is ambiguous, follow up with a more specific typed query to disambiguate.

Working with multiple Revit sessions

Archi Automate can detect and switch between several running Revit instances, so one AI client can work across multiple concurrent projects. When more than one session is open, you choose which one the AI operates on.

1. Ask the AI to list available Revit sessions. It calls revit_list_sessions and presents the running sessions it discovered.
2. Tell the AI which session to use. It calls revit_select_session to pin that session as active.
3. All subsequent operations — document queries, element collection, read operations, and write tasks — route to the selected session automatically.
4. You can switch the active session at any time by repeating the selection step.

If only one Revit instance is running, selection is automatic and you can go straight to asking questions. A session must be selected before document or element operations will work; if the AI reports it cannot find a session, confirm that Revit is open with the Bridge loaded in the target instance.

MCP tool reference

The table below lists the MCP tools that Archi Automate exposes to a connected AI client. In normal use you do not call these directly — you ask in natural language and the AI selects the appropriate tool — but the reference is useful for understanding what happens behind each request. Read-tier tools are always permitted; write-tier tools require write mode to be enabled in policy.

Tool	Tier	What it does
revit_list_sessions	Read	Discovers all running Revit sessions so the AI can route requests to the right one.
revit_select_session	Read	Pins the active session; all subsequent tool calls are routed to it.
revit_ping	Read	Verifies that the Bridge in the selected session is alive and responding.
revit_list_documents	Read	Lists all documents currently open in the selected session.
revit_get_document_info	Read	Returns metadata (title, path, version, project information) for the active or a named document.
revit_collect_elements	Read	Collects elements by category, class, or filter criteria, returning matching element IDs with basic metadata.
revit_get_parameters	Read	Retrieves current parameter values for a set of element IDs.
revit_run_readonly_operation	Read (advanced)	Carries out an advanced read-only operation across the full Revit API, for queries the structured tools cannot express.
revit_run_write_task	Write (advanced)	Carries out an advanced write operation inside a managed Revit transaction, with automatic rollback on error. Requires write mode in policy.
revit_activate_license	Admin	Activates or checks the current license state. This is an administrator-tier tool and runs in an elevated context, not part of everyday user workflows.

Guardrails and safety

Archi Automate is built so that AI access to your Revit model can be deployed safely across a team. Every request passes through a layered set of guardrails before anything runs: an execution-mode gate decides whether write operations are allowed at all, an operation-safety check blocks dangerous actions before they run, structured scope limits cap how much of the model a single operation can touch, and an audit trail records everything the AI does. These layers work together as defense-in-depth, not as a single barrier.

Execution modes

The execution mode is the master control over whether the AI can change the model. It is set by the executionMode policy field and applies to every connected session governed by that policy. There are three modes:

• readOnly (default) — Only read-tier tools are permitted. The AI can list sessions, query documents, collect elements, retrieve parameters, and run advanced read-only operations. Any write task is rejected immediately with a not_allowed error. This is the safe starting point and the right mode for users who only need to query and analyze the model.
• dryRun (preview) — Write tasks are allowed to run, but the surrounding Revit transaction is always rolled back when the task completes. The AI carries out the operation so you can see exactly what it would change, but nothing is committed to the model. Use this to validate that the AI's approach is correct before allowing real changes.
• unrestricted (commit) — Write tasks run inside a managed transaction and commit on success. This is full model-modification access and should be enabled deliberately, ideally only after prompting patterns have been validated in dryRun.

Recommended production default: run teams in dryRun until your prompting patterns are proven, then move to unrestricted for the users and workflows that need to commit changes. Promotion to unrestricted is best gated behind an explicit administrator decision in the company-wide policy layer.

Note that even in dryRun mode, side effects that occur outside the Revit transaction (for example UI notifications) are not rolled back. The rollback guarantee covers model changes made within the transaction.

Operation safety

Every operation is reviewed by a safety layer before it runs. The safety layer blocks dangerous actions that have no place in a Revit task, including:

• File and disk access
• Network access
• Reaching outside the Revit environment
• Background threading
• Launching other programs

If an operation attempts a blocked action, it is rejected before anything runs. The check is controlled by the codeSafety.enabled field and can be extended with organization-specific restrictions through codeSafety.additionalDeniedNamespaces.

Important: the safety layer is a defense-in-depth mechanism, not a sandbox. It blocks the most dangerous actions, but it does not isolate execution and does not guarantee that every incorrect or undesirable operation is prevented. Always combine operation safety with an appropriate execution mode and scope limits, and preview unfamiliar operations in dryRun first.

Scope limits

Structured scope settings constrain how much of the model a single operation can address, independent of the execution mode. They are configured under the scope section of the policy:

• Maximum elements per operation (scope.maxElementsPerOperation) — caps the number of elements a single operation can address. Element collection and operation results are truncated or rejected when the limit is exceeded, protecting Revit from being overwhelmed by an over-broad query. The default is 5000; a value of 0 removes the limit.
• Category allow / deny lists (scope.allowedCategories and scope.deniedCategories) — optional lists of Revit built-in category names that restrict which parts of the model the AI can access or modify. Both are empty by default, meaning no category restriction.
• Block-deletes guard (scope.blockDeletes) — when set, write tasks that call Document.Delete are rejected even in unrestricted mode. This is a hard guard against accidental element deletion and is enabled by default.

Audit trail

When auditing is enabled (audit.enabled, on by default), every request the AI makes is logged as a JSONL entry under the local audit folder. Each entry captures the request and its outcome, giving administrators a complete, machine-readable record of all AI activity against the model — what was asked, what ran, and what resulted.

Audit logs remain local to the machine unless an administrator configures a remote log sink through the company-wide policy. The audit folder is preserved across uninstall so historical records are not lost.

The Hub dashboard

The Hub is a standalone WPF application installed alongside Archi Automate. It is version-independent and manages all installed Revit versions from a single window. It serves two main purposes:

• Live monitoring — the Hub displays the current system status and a list of connected Revit sessions, auto-refreshing as sessions come and go. Use it to confirm which sessions the AI can reach. The Hub needs at least one running Bridge to show live session data.
• Visual policy editing — the Guardrails view exposes every policy field as a form, so you can adjust the execution mode, operation-safety settings, and scope limits without editing JSON by hand.

Policy changes saved from the Hub are written to the local policy file and take effect on the next Bridge startup or policy reload. Restart the Bridge in the affected Revit sessions for a change to apply.

Settings reference

The policy file

All guardrail settings are stored in a single local JSON policy file. It is created automatically with safe defaults the first time the Bridge starts, so no manual setup is required to run securely. Every field in the file corresponds to the guardrail policy schema shared by the Bridge and MCP server.

You can edit the policy in two ways: directly in the local policy file, or — for most fields — through the Hub's Guardrails view, which is the recommended approach for day-to-day administration. In either case, changes apply on the next Bridge startup or policy reload.

An optional company-wide policy layer can override or supplement the local file across a team. This layer is fail-closed: if a required company policy cannot be fetched (for example when the machine is offline), the Bridge falls back to the most restrictive settings rather than failing open. The company layer is fetched at Bridge startup, so changes made to it take effect only after the Bridge reloads.

Policy fields

The table below lists the guardrail policy fields, their types, defaults, and effect. Fields marked as editable in the Hub can also be changed from the Guardrails view; all fields can be set in the local policy file.

Setting	Type	Default	What it does
executionMode	Enum (readOnly, dryRun, unrestricted)	readOnly	Controls whether the AI can modify the model. readOnly rejects all write tasks; dryRun runs write tasks but rolls back the transaction for preview; unrestricted commits write tasks.
codeSafety.enabled	Boolean	true	Turns the operation-safety checks on or off. When enabled, dangerous actions such as file, network, and process access are blocked before any operation runs.
codeSafety.additionalDeniedNamespaces	String array	empty	Extends the safety blocklist with additional restricted API areas defined by your organization.
scope.maxElementsPerOperation	Integer	5000	Caps the number of elements a single operation can address. Results are truncated or rejected when exceeded. A value of 0 means no limit.
scope.blockDeletes	Boolean	true	Rejects write tasks that call Document.Delete — even in unrestricted mode — as a hard guard against accidental element deletion.
scope.allowedCategories	String array	empty	Restricts AI access and modification to the listed Revit built-in categories. Empty means no restriction.
scope.deniedCategories	String array	empty	Excludes the listed Revit built-in categories from AI access and modification. Empty means no exclusion.
audit.enabled	Boolean	true	Logs every request and its outcome as JSONL in the local audit folder, producing a full audit trail of AI activity against the model.
watchdogTimeoutSeconds	Integer	60	Maximum runtime for a single read or write operation. Operations that exceed it are cancelled and return a timeout error, keeping Revit responsive.

Workflows

Each workflow below is goal-based: it states when to use it and the steps to follow. All workflows assume Revit is open with a document loaded, the Archi Automate Bridge is loaded, and an MCP-compatible AI client (Claude Desktop, Claude Code, or a compatible GPT/Codex agent) is connected.

Query the model without knowing Revit

Use this when you want information from a Revit project without knowing Revit commands or the Revit API.

1. Open the AI client with Archi Automate connected.
2. Ask in plain language, for example: How many walls are load-bearing? or List all rooms larger than 20 square meters.
3. The AI calls the element collection and parameter retrieval tools, or carries out an advanced read-only operation if the question needs custom logic.
4. The answer is returned directly from the live model and explained in natural language.

If the AI reports it cannot find a session, ask it to list and select sessions first. If a query is rejected for scope, narrow the filter so it stays under the maxElementsPerOperation limit.

Automate a repetitive parameter update

Use this when you need to update a parameter across many elements without editing each one by hand.

1. Describe the rule, for example: Set the Fire Rating parameter to '60 min' for all walls in the Core zone.
2. The AI prepares a write task targeting the matching elements.
3. Review the planned changes in dryRun mode, where the transaction is rolled back so nothing is committed.
4. Approve the operation, then execute it in unrestricted mode so the transaction commits.
5. The AI returns a summary of the changed elements. Save the model manually when you are satisfied.

Write mode must be enabled in policy. If the task is rejected as not_allowed, the active execution mode is readOnly; adjust it in the Hub first.

Run a model audit

Use this when you want to check the model for compliance issues, missing data, or naming inconsistencies.

1. Describe the audit, for example: Find all rooms with no department parameter set.
2. The AI carries out an advanced read-only operation.
3. The result is a list of element IDs and locations for you to review.
4. Follow up with a more specific question or a corrective write task if you want to fix the findings.

Audits are read-only and never modify the model, so they are safe to run at any time and in any execution mode.

Compose a complex multi-category operation

Use this when a task spans multiple element categories, relationships, and conditions that no fixed tool could express.

1. Describe the operation in as much detail as needed, including the categories, conditions, and the result you expect.
2. The AI refines the operation until it succeeds, adjusting its approach if Revit reports a problem.
3. Preview the operation in dryRun mode before committing anything unfamiliar.
4. The operation executes inside a managed Revit transaction; if anything fails or Revit rejects an operation, the transaction is rolled back automatically and the model is left unchanged.
5. The AI returns a structured summary of the changed elements.

Every operation passes a defense-in-depth safety layer before it runs, but it is not sandboxed. Always preview unfamiliar operations with dryRun first.

Switch between Revit sessions

Use this when several Revit instances are open and you want to direct the AI at a specific project.

1. Ask the AI: List available Revit sessions. It calls revit_list_sessions and presents the discovered sessions.
2. Tell the AI which session to use.
3. The AI calls revit_select_session to pin that session.
4. All subsequent queries and tasks operate against the selected session.

The Bridge must be loaded in the target session. If a session does not appear, confirm Revit is running there with the Bridge loaded, and restart that instance if needed.

Review and adjust guardrail policy

Use this when an administrator wants to change what the AI is allowed to do.

1. Open the Hub and go to the Guardrails view.
2. Adjust the execution mode (readOnly, dryRun, or unrestricted), the operation-safety settings, or the scope limits such as maxElementsPerOperation and blockDeletes.
3. Save the policy.
4. Restart the Bridge in the affected Revit sessions so the change takes effect.

A company-wide policy layer can override local settings. If a change does not take effect, contact your IT administrator to review the company policy.

Troubleshooting

Archi Automate does not load in Revit

Symptoms: No Bridge pipe is available and the AI client reports a connection failure.

Likely cause: The installation is incomplete or used the wrong Bridge build for the current Revit version, the add-in manifest is missing or points to the wrong path, or there is a .NET runtime mismatch.

Fix: Reinstall Archi Automate with administrator rights and make sure Revit is closed during installation. Verify that the ArchiAutomate.Bridge.addin manifest is present for the active Revit version. Bridge startup errors are recorded under %LocalAppData%\ArchiAutomate\logs\.

The AI cannot find any Revit sessions

Symptoms: revit_list_sessions returns an empty list, and the Hub shows no connected sessions.

Likely cause: Revit is not running, or the Bridge is not loaded in the running Revit instance.

Fix: Confirm that Revit is open with a document loaded and the Bridge is registered for that Revit version. Restart Revit so the Bridge loads fresh; the Hub auto-refreshes and will detect the session.

A write task is rejected as not allowed

Symptoms: The AI reports the operation is not permitted with a not_allowed error.

Likely cause: The executionMode is set to readOnly, a company policy overrides the local mode, blockDeletes is active and the operation attempts a delete, or the target category is in deniedCategories.

Fix: Open the Hub and check the current execution mode and scope settings. Change executionMode to dryRun or unrestricted as appropriate. If a company policy is blocking the operation, contact your IT administrator.

An operation could not be completed

Symptoms: The AI reports that it could not carry out the requested operation.

Likely cause: The request was ambiguous or could not be expressed against the available Revit API for that version, or it attempted an action blocked by the safety layer.

Fix: Ask the AI to rephrase or retry, or describe the request more specifically. If the safety layer blocked the action, ask the AI to use an alternative approach that does not require the restricted action.

An operation times out

Symptoms: The AI receives a timeout response, and Revit may have been briefly unresponsive during the operation.

Likely cause: The operation ran a very large query or loop, or Revit was busy with another task while it ran.

Fix: Ask the AI to narrow the query scope or process elements in smaller batches. If the workload is genuinely large, raise watchdogTimeoutSeconds in policy. The Revit session remains usable after a timeout.

A write task failed and rolled back

Symptoms: The AI receives a transaction_failed response and the model is unchanged.

Likely cause: The operation failed, or Revit rejected it (for example, a read-only element or the wrong element type).

Fix: The transaction is rolled back automatically, so model integrity is preserved. Ask the AI to revise the write task based on the error detail. If an element is read-only or managed by a workset, check model permissions before retrying.

Frequently asked questions

Q. Do I need an AI client to use Archi Automate?
A. Yes. Archi Automate requires a connected MCP-compatible AI client, such as Claude Desktop, Claude Code, or a compatible GPT/Codex agent. The Bridge runs silently inside Revit and has no ribbon button; all interaction happens through the AI client.

Q. Which AI models work best?
A. Result quality depends heavily on the model tier. The recommended minimums for reliable results are Claude Opus 4.8 High or better, and GPT-5.5 High or better.

Q. Does Archi Automate modify my model automatically?
A. No. Write mode must be explicitly enabled in policy; in the default readOnly mode write tasks are rejected. Use dryRun mode to preview changes, and note that all writes run inside a managed Revit transaction with automatic rollback on error.

Q. Is it safe to let the AI make changes?
A. Archi Automate is read-only by default, every operation passes a defense-in-depth safety layer that blocks dangerous actions such as file, network, and process access, and all write operations run inside a managed Revit transaction with automatic rollback plus a full audit trail. This is defense-in-depth, not a sandbox, so always preview unfamiliar operations with dryRun first.

Q. Which Revit versions are supported?
A. Revit 2025, 2026, and 2027.

Q. Does Archi Automate send my model to the cloud?
A. No. Revit model data is not transmitted to external services. The Bridge and the MCP server communicate exclusively over a local pipe on the same machine.

Q. Does it save my model after a write task?
A. No. Archi Automate does not save the Revit model after a write task completes. You save manually once you are satisfied with the result.